Hey, I'm Damon Millar. I'm an accountant, fractional CFO, and business advisor. I help business owners grow profits, reduce tax, and build lasting wealth, and I help accounting firms run a more proactive, higher-value practice.
I built this bundle inside my own firm before I wrote a word of it. It solves a narrow problem. Everyone in practice knows AI is useful by now, and almost nobody has it set up as a system. So it drafts the odd email, saves twenty minutes, and never touches the work that eats the week.
This is the build layer. Seven things you can construct in an afternoon, each one a permanent part of how the firm runs rather than a prompt you retype.
Everything here is written for Claude specifically, because Projects, Skills and scheduled tasks are what turn it into a system.
<aside>
📂 I've built everything in this bundle inside my own practice first. If you want a second set of eyes on yours, a 30-minute conversation will identify where this would create the most useful capacity.
Book a call → https://calendly.com/damon-millar-switchaccountants/30min
</aside>
Nothing below works without these. They take five minutes and they are the difference between a firm that uses AI safely and a firm that has an incident.
<aside> ⚠️
Client data comes first. On Claude Free or Pro, turn off model training in Settings → Privacy. Better: use Team or Enterprise, where your data isn't used for training by default. Anonymise before you paste. "Client A, engineering, £1.4m turnover, 3 shareholders" carries every bit of context the model needs. No names, no UTRs, no NI numbers, no bank details.
</aside>
1. It drafts. A qualified human signs. Nothing tax-technical or client-facing leaves the building unreviewed. Your name is on the output and your PII cover is behind it.
2. Every rate, threshold and allowance is a prompt to check, not an answer. Model training data goes stale, and UK thresholds move. Treat any figure it gives you as a pointer to the current HMRC guidance, never as the figure itself. Every prompt in this bundle is written to flag rather than state.
3. Context beats cleverness. A plain prompt loaded with your firm's real templates, tone and the client's actual numbers beats a clever prompt with none. That's why Resource 1 comes first. Skip it and the other six underperform.
4. Save what works, immediately. The moment a prompt gives you a good result, it goes in the Project or becomes a Skill. A prompt you rewrite from scratch each time isn't a system.
5. Measure the hours back. If a build doesn't return real time or real quality, drop it. Busy-looking AI use is still busywork.
Build Resource 1 today. It takes about ten minutes and everything else depends on it.
Then pick the one of the remaining six that hits your biggest time-sink and build only that. Use it for two weeks until it's habit. Then come back for the next.
Firms that try to build all seven in a weekend end up with seven half-configured things nobody trusts.